Definition
Seasonal differencing subtracts an observation one seasonal cycle earlier from the current observation. With daily data and weekly seasonality, the difference is today minus seven days earlier. It can help stabilize recurring seasonal level patterns before fitting a time-series model, but it is not a universal requirement or a replacement for demand-quality checks.
Why It Matters
- Repeated weekly patterns can obscure changes in the demand level that a replenishment model needs to learn.
- The Commerce Intelligence OS framework should preserve both the original sales evidence and the transformation used by a predictive workflow so buying decisions remain explainable.
How It Works
- Establish a regular time grid and a defensible seasonal period. Distinguish missing observations from confirmed zero sales and identify periods constrained by unavailable inventory.
- Assess whether seasonal differencing is appropriate for the selected model. Inspect the series and diagnostics instead of applying it automatically to every SKU.
- Fit and evaluate the transformed model with historical cutoffs respected. Avoid unnecessary additional differencing, which can introduce artificial patterns.
- Convert forecasts back to original units using the required historical or recursively forecast values. Evaluate the actual planning horizon in units and commercial consequences.
Ecommerce Example
Context: Illustrative example: a store records 40 units this Monday and 34 units on the previous Monday. The weekly seasonal difference is six units.
Recommended move: A model predicts next Monday will exceed this Monday by four units. The reconstructed point forecast is 44 units, assuming that is the transformation and model output being used.
Why it matters: The buyer still checks promotions and availability. A difference of four is a change, not a four-unit demand forecast; these figures are illustrative.
iKawn Framework
Preserve
The iKawn framework retains original demand observations and availability context.
Transform
Version the seasonal period and transformation rules.
Reconstruct
Return model output to the units used by inventory teams.
Evaluate
Compare future-period planning accuracy before adopting the transformation.
Concise Summary
Seasonal differencing compares matching positions in successive cycles. Apply it only when appropriate, preserve original observations, and reconstruct forecasts before operational use.